Constant high omega for few particles close to base of system

msbentley's picture
Submitted by msbentley on Thu, 08/16/2012 - 15:13

This may be a bug, or just a lack in my understanding - but I have noticed in many sims that often get one or two particles with unexpectedly high omega, usually at the bottom of the sim domain (in contact with the lower wall). As a simple example, I took the in.newModels example from the contactModels tutorial and simply ran for a bit longer (150000 instead of 50000 timesteps).

Assuming that omega is here in rad/s, a constant value of ~2 rad/s seems quite high. In the attachments you can see a plot of the energies, a histogram of the distribution for all particles at the last timestep, and a plot of the magnitude and individual components of omega for this anomalous particle over time.

Is this typical? Or just a function of the non-realistic parameters used in this example? I looked at this for a few different sims, and in all cases the z-component of omega for a few particles in contact with the floor seemed high..

I'm running LIGGGHTS 2.0.6. Thanks for your hints!

Mark

ckloss's picture

ckloss | Mon, 08/20/2012 - 23:01

All the standard examples work at the stability limit with unrealistically low Young's Modulus - so I would not worry too much about that. Duration of a translational contact is currently not tracked in fix check/timestep/gran - most probably the time-step is simply too high to capture roatational dynamics accurately

Cheers, Christoph

msbentley's picture

msbentley | Mon, 08/27/2012 - 14:20

Hi Christoph,

That was my assumption at first, but I re-ran the same case with a Young's modulus of 5.e8 and a timestep of 1.e-8 s - after a pretty long run of 3.5e8 timesteps I still see the same - one or two particles with unexpectedly high omega, almost entirely in the z component:

It probably doesn't make any different to the large scale dynamics of the system, but I'd like to track down the reason - any hints gratefully accepted!

Thanks, Mark

ckloss's picture

ckloss | Mon, 08/27/2012 - 16:32

Hi mark,

plotting the coordination number could maybe help. I guess these particles simply rest at the wall, with no other particle contacts so they end up spinning in the z axis forever since there is no torque opposed to this spinning movement in standard DEM

Christoph

msbentley's picture

msbentley | Tue, 08/28/2012 - 12:38

Yes, I'll do that - it was this possibility that led me to enquire about coord/gran in the old version - now it's included again I'll give it a try! The force on these particles is indeed lower than some surrounding ones.

Thanks for your help!

Edit: yep, looks like the culprit - all of the particles with high omegas have a coordination number of zero - I assume compute contact/atom only includes particle interactions, and not particle-wall? So these particles are "spinning tops", having high rates around the Z axis and only being in contact with the floor:

cstoltz | Tue, 08/28/2012 - 12:53

Mark,

Thanks for the update, and great visuals. One question - is it physically sensible to have isolated particles along the bottom layer of a set of particles inside a container? You would need the surrounding particles to effectively make a shell surrounding the isolated 'spinning' one, which doesn't seem too likely to me.

Any thoughts?

Thanks - Chris

ckloss's picture

ckloss | Tue, 08/28/2012 - 15:16

Hi Mark and Chris,

this example is just 4 or 5 layers of particles, so it doesn't seem unlikely to me.

Cheers,
Christoph

richti83's picture

richti83 | Fri, 08/31/2012 - 14:48

Hi all,

I'm reading this thread for a while and want to share/discuss the rolling friction model type here with you:
First I injected the contactmodel at gran/hertz/history where torque_r[] is calculated to print out this value to file, with the original code we got a constant value with alternating sign every step. The reason is the w/w_mag function which should give the sign of r_troque. Our Phd H. Otto suggested to use a tanh(w*factor) function to soften the zero-crossing of the sign function.
Et voila:

(one particle v0_x=1m/s rolling with friction on a wall)

It's only a starting point, our goal is to implement modeltype C (Ai et.all), but in my opinion the spinning particle results of the left image ..

Have a nice weekend,
Christian

I'm not an associate of DCS GmbH and not a core developer of LIGGGHTS®
ResearchGate | Contact

msbentley's picture

msbentley | Fri, 08/31/2012 - 13:13

... of a system with 188k particles:

where we are looking at the base of the system - again the particles with high omega are those with a coordination number of zero. There are 54 particles here with zero coordination number. Unfortunately I didn't do a dump local at the end of this run, but I'll try a restart and do that.

It still looks a little odd to me - any further comments or suggestions welcome!

Cheers, Mark

msbentley's picture

msbentley | Tue, 08/28/2012 - 15:20

Good question - they are not supposed to be isolated - this is again a modified version of the contact model example (in.newModels), so particles are inserted and fall under gravity. I suppose it's possible that somehow there is a static arch here "protecting" these particles; I'm not sure how likely it is this would happen ever time, though...

In some of my recent sims I see much more pronounced examples of this - particles always close to the boundary with ~300 rad/s angular velocity etc.

I'll try some other cases - my "real" runs are using much smaller particles (few hundred microns) and CGS units - I thought some of the odd behaviour might be my choice of skin depth for the neighbour list, but I will have to try varying a few parameters to see if I always get this behaviour.

Cheers, Mark